AMD Enables High-Performance Video Generation with Day 0 Support for MiniMax-H3
AMD has announced Day 0 support for MiniMax-H3, a cutting-edge video generation model from Chinese AI firm MiniMax. This model, released on July 31, 2026, represents a generational leap in video synthesis and can generate up to 15-second clips at 2K resolution (1440p) with synchronized stereo audio.
The support relies on AMD's ROCm platform and SGLang framework to optimize MiniMax-H3's compute-heavy architecture. By employing 8-way sequence parallelism and leveraging AMD’s AI Tensor Engine (AITER) kernels, the GPUs ensure smooth processing of the joint video-audio latents that form the core of H3’s output.
MiniMax-H3 is an open-weight model available under the MiniMax Community License, making it accessible to developers for experimentation. The model is already available on platforms like Atlas, where users can access it for $0.14 per second at 2K resolution.